Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Proposition 8 — unique backup-cost threshold and primary-cost derivative

Disproved
WangWebsterFlexibility.Threshold.proposition_8

by Shuze Chen · Oct 29, 2026 · Mathlib 0df444a (Lean v4.33.1)

operations-researchp2o-gran-per-chapterp2o-plan-paperp2o-v1stochastic-optimizationsupply-chain

Fix arbitrary independent nonnegative demand laws, margins π1>π2\pi_1>\pi_2π1​>π2​, failure probability 0<ρ<10<\rho<10<ρ<1, and exercise cost 0≤v<π20\le v<\pi_20≤v<π2​. There exists a real-valued threshold function Γ(c)\Gamma(c)Γ(c) on 0≤c≤v0\le c\le v0≤c≤v. At each such ccc, its value is the unique finite real number satisfying all four threshold clauses: at least one admissible fee lies below or at the cutoff; df weakly dominates ff for every such fee; ff weakly dominates df for every admissible fee above the cutoff; and strict ff dominance occurs for at least one fee above it. Admissibility means γ≥0\gamma\ge0γ≥0 and γ+v>c\gamma+v>cγ+v>c; the two networks optimize their reservations and primary orders independently.

For every 0<c<v0<c<v0<c<v, a finite derivative exists and

Γ′(c)≤0.\Gamma'(c)\le0.Γ′(c)≤0.

At c=0c=0c=0 and c=vc=vc=v a derivative within [0,v][0,v][0,v] exists with a nonpositive value. The function's extension outside [0,v][0,v][0,v] is unconstrained. For v=0v=0v=0 the interior clause is empty and the within-singleton derivative condition is vacuous; the unique threshold and all dominance clauses at c=0c=0c=0 remain required. No differentiability, bounded demand, unique optimizer, or single crossing is assumed.

This is an open statement of the full source result under the declared strong reading, not a machine-checked proof. Insight I2 records the source proof gap and insight I3 records failure of its strict-dominance clause for zero demands under the fixed-law reading.

Differences from the source:

  • encoding: independent nonnegative real random demands and nonnegative real decisions → Borel probability laws on nonnegative reals, their product measure, and nonnegative-real decision types; reason: the law-based representation covers all the original distributions and decisions, including atoms and unbounded support
  • encoding: separately optimized network profits → extended-real suprema over all feasible finite reservations and orders; reason: no boundedness or attainment assumption is supplied by the source; positive infinity must not become a default real zero
  • reading: “for some γ+v≤Γ” on p. 21, versus “for any γ+v≤Γ” and the reverse weak inequality on A16 → all four strong single-crossing clauses with both existential witnesses, uniqueness at each c, fixed demand laws and fixed v while γ varies; reason: the weaker two-witness reading cannot characterize a unique cutoff; the appendix supplies the universal regime comparisons
  • encoding: ∂Γ/∂c≤0 on the admissible cost domain → existence of a nonpositive ordinary derivative for 0<c<v and a nonpositive derivative within [0,v] at its endpoints; reason: the domain has boundary points and a total derivative default must not substitute for derivative existence; at v=0 only the singleton boundary convention applies
Preamble
import Definitions.Def_WangWebsterFlexibility_Threshold_Cutoff
import Mathlib.Analysis.Calculus.Deriv.Basic

open scoped NNReal
Formal statement
/- Imported definition bodies for the blind statement audit.
These are quotations of the imported Lean modules, not new declarations.

import Mathlib.MeasureTheory.Measure.ProbabilityMeasure
import Mathlib.MeasureTheory.Measure.Prod
import Mathlib.MeasureTheory.Integral.Bochner.Basic

open MeasureTheory

open scoped NNReal

namespace WangWebsterFlexibility.Threshold

/-- Two independent nonnegative demand laws and the common economic primitives (§3, pp. 9–11).
The laws are unrestricted Borel probability measures; no moment or density is assumed. -/
structure Model where
  demand₁ : ProbabilityMeasure ℝ≥0
  demand₂ : ProbabilityMeasure ℝ≥0
  π₁ : ℝ
  π₂ : ℝ
  ρ : ℝ

/-- Independence is implemented with the standard product measure. -/
noncomputable def Model.law (M : Model) : Measure (ℝ≥0 × ℝ≥0) :=
  (M.demand₁ : Measure ℝ≥0).prod (M.demand₂ : Measure ℝ≥0)

/-- Expected value under the independent demand laws. The sales expressions used below are
measurable and bounded for each fixed finite decision, even when demand has infinite mean. -/
noncomputable def Model.expect (M : Model) (f : ℝ≥0 × ℝ≥0 → ℝ) : ℝ :=
  ∫ d, f d ∂M.law

/-- Real positive part, used in the printed allocation formulas. -/
noncomputable def pos (x : ℝ) : ℝ := max 0 x

/-- Revenue net of exercise charges with dedicated available primary quantities (4), p. 12. -/
noncomputable def revenueDF (M : Model) (v : ℝ) (d : ℝ≥0 × ℝ≥0)
    (a₁ a₂ s : ℝ) : ℝ :=
  M.π₁ * min (d.1 : ℝ) a₁ + M.π₂ * min (d.2 : ℝ) a₂ +
    (M.π₁ - v) * min (pos ((d.1 : ℝ) - a₁)) s +
    (M.π₂ - v) * min (pos ((d.2 : ℝ) - a₂)) (pos (s - pos ((d.1 : ℝ) - a₁)))

/-- Revenue net of exercise charges when the flexible primary succeeds (5), p. 13. -/
noncomputable def revenueFF (M : Model) (v : ℝ) (d : ℝ≥0 × ℝ≥0)
    (q s : ℝ) : ℝ :=
  M.π₁ * min (d.1 : ℝ) q + M.π₂ * min (d.2 : ℝ) (pos (q - (d.1 : ℝ))) +
    (M.π₁ - v) * min (pos ((d.1 : ℝ) - q)) s +
    (M.π₂ - v) * min (pos ((d.2 : ℝ) - pos (q - (d.1 : ℝ))))
      (pos (s - pos ((d.1 : ℝ) - q)))

end WangWebsterFlexibility.Threshold


import Definitions.Def_WangWebsterFlexibility_Threshold_Model
import Mathlib.Data.EReal.Basic

open scoped NNReal

namespace WangWebsterFlexibility.Threshold

/-- Equation (4): independent primary failures, reservation paid up front, primary paid
on delivery including unsold units. Decisions are all nonnegative finite real numbers. -/
noncomputable def profitDF (M : Model) (c γ v : ℝ) (s q₁ q₂ : ℝ≥0) : ℝ :=
  -γ * (s : ℝ) - c * (1 - M.ρ) * ((q₁ : ℝ) + (q₂ : ℝ)) +
    M.ρ ^ 2 * M.expect (fun d => revenueDF M v d 0 0 s) +
    (1 - M.ρ) ^ 2 * M.expect (fun d => revenueDF M v d q₁ q₂ s) +
    M.ρ * (1 - M.ρ) * M.expect (fun d => revenueDF M v d q₁ 0 s) +
    M.ρ * (1 - M.ρ) * M.expect (fun d => revenueDF M v d 0 q₂ s)

/-- Equation (5): one common failure of the consolidated primary supplier. -/
noncomputable def profitFF (M : Model) (c γ v : ℝ) (s q : ℝ≥0) : ℝ :=
  -γ * (s : ℝ) - c * (1 - M.ρ) * (q : ℝ) +
    M.ρ * M.expect (fun d => revenueDF M v d 0 0 s) +
    (1 - M.ρ) * M.expect (fun d => revenueFF M v d q s)

/-- Independently optimized df value over the full feasible decision set, including zero.
The complete extended real order retains an unbounded supremum as positive infinity. -/
noncomputable def valueDF (M : Model) (c γ v : ℝ) : EReal :=
  ⨆ (s : ℝ≥0) (q₁ : ℝ≥0) (q₂ : ℝ≥0), (profitDF M c γ v s q₁ q₂ : EReal)

/-- Independently optimized ff value, with no assumed maximizer. -/
noncomputable def valueFF (M : Model) (c γ v : ℝ) : EReal :=
  ⨆ (s : ℝ≥0) (q : ℝ≥0), (profitFF M c γ v s q : EReal)

/-- Admissible reservation fee at fixed primary and exercise costs (§3, pp. 10–11). -/
def AdmissibleFee (c v γ : ℝ) : Prop := 0 ≤ γ ∧ c < γ + v

end WangWebsterFlexibility.Threshold


import Definitions.Def_WangWebsterFlexibility_Threshold_Profits

open scoped NNReal

namespace WangWebsterFlexibility.Threshold

/-- The strong single-crossing reading of Proposition 8(a) (p. 21 and A16).
Existence, uniqueness and regularity are deliberately not fields of the model. -/
def IsThreshold (M : Model) (c v Γ : ℝ) : Prop :=
  (∃ γ : ℝ, AdmissibleFee c v γ ∧ γ + v ≤ Γ) ∧
  (∀ γ : ℝ, AdmissibleFee c v γ → γ + v ≤ Γ →
    valueFF M c γ v ≤ valueDF M c γ v) ∧
  (∀ γ : ℝ, AdmissibleFee c v γ → Γ < γ + v →
    valueDF M c γ v ≤ valueFF M c γ v) ∧
  (∃ γ : ℝ, AdmissibleFee c v γ ∧ Γ < γ + v ∧
    valueDF M c γ v < valueFF M c γ v)

end WangWebsterFlexibility.Threshold

-/

namespace WangWebsterFlexibility.Threshold

/-- Proposition 8, p. 21, all clauses; the reading of some/any follows A16.
Ordinary derivatives are required inside [0,v], within-interval derivatives at its endpoints. -/
theorem proposition_8 (M : Model)
    (hπ : M.π₂ < M.π₁)
    (hρ₀ : 0 < M.ρ) (hρ₁ : M.ρ < 1)
    (v : ℝ) (hv : 0 ≤ v) (hvπ : v < M.π₂) :
    ∃ Γ : ℝ → ℝ,
      (∀ c ∈ Set.Icc 0 v, IsThreshold M c v (Γ c) ∧
        ∀ G : ℝ, IsThreshold M c v G → G = Γ c) ∧
      (∀ c ∈ Set.Ioo 0 v, ∃ a : ℝ, a ≤ 0 ∧ HasDerivAt Γ a c) ∧
      (∀ c : ℝ, c = 0 ∨ c = v →
        ∃ a : ℝ, a ≤ 0 ∧ HasDerivWithinAt Γ a (Set.Icc 0 v) c) := by sorry

end WangWebsterFlexibility.Threshold
Source
Wang and Webster (2021), Product Flexibility Strategy Under Supply and Demand Risk, Manufacturing & Service Operations Management, https://doi.org/10.1287/msom.2021.1037, Proposition 8(a)–(b), printed/PDF p. 21; proof Appendix A2, printed A11–A16 / PDF 48–53
Read-back

What the Lean code literally says, in plain math · claude-opus-5-5

Setting. The statement concerns a model MMM with the following data:

  • two Borel probability measures μ1,μ2\mu_1, \mu_2μ1​,μ2​ on the nonnegative reals [0,∞)[0,\infty)[0,∞), with no moment, density or other condition imposed;
  • three real numbers π1,π2,ρ\pi_1, \pi_2, \rhoπ1​,π2​,ρ.

Write D=(D1,D2)D = (D_1, D_2)D=(D1​,D2​) for a point of [0,∞)2[0,\infty)^2[0,∞)2, distributed according to the product measure μ1⊗μ2\mu_1 \otimes \mu_2μ1​⊗μ2​. For a real function fff of DDD,

E[f]=∫f(D) d(μ1⊗μ2)(D),\mathbb E[f] = \int f(D)\, d(\mu_1\otimes\mu_2)(D),E[f]=∫f(D)d(μ1​⊗μ2​)(D),

with the convention that the integral is 000 when fff is not integrable. Throughout, x+=max⁡(0,x)x^+ = \max(0,x)x+=max(0,x).

Revenue functions. For an exercise cost vvv and real numbers a1,a2,s,qa_1, a_2, s, qa1​,a2​,s,q, define

RDF(D;a1,a2,s)=π1min⁡(D1,a1)+π2min⁡(D2,a2)+(π1−v)min⁡((D1−a1)+, s)+(π2−v)min⁡((D2−a2)+, (s−(D1−a1)+)+),R^{DF}(D;a_1,a_2,s) = \pi_1\min(D_1,a_1) + \pi_2\min(D_2,a_2) + (\pi_1-v)\min\big((D_1-a_1)^+,\, s\big) + (\pi_2-v)\min\big((D_2-a_2)^+,\, (s-(D_1-a_1)^+)^+\big),RDF(D;a1​,a2​,s)=π1​min(D1​,a1​)+π2​min(D2​,a2​)+(π1​−v)min((D1​−a1​)+,s)+(π2​−v)min((D2​−a2​)+,(s−(D1​−a1​)+)+), RFF(D;q,s)=π1min⁡(D1,q)+π2min⁡(D2,(q−D1)+)+(π1−v)min⁡((D1−q)+, s)+(π2−v)min⁡((D2−(q−D1)+)+, (s−(D1−q)+)+).R^{FF}(D;q,s) = \pi_1\min(D_1,q) + \pi_2\min\big(D_2,(q-D_1)^+\big) + (\pi_1-v)\min\big((D_1-q)^+,\, s\big) + (\pi_2-v)\min\Big(\big(D_2-(q-D_1)^+\big)^+,\, \big(s-(D_1-q)^+\big)^+\Big).RFF(D;q,s)=π1​min(D1​,q)+π2​min(D2​,(q−D1​)+)+(π1​−v)min((D1​−q)+,s)+(π2​−v)min((D2​−(q−D1​)+)+,(s−(D1​−q)+)+).

Profit functions. For real c,γ,vc, \gamma, vc,γ,v and nonnegative reals s,q1,q2,qs, q_1, q_2, qs,q1​,q2​,q, define

Pc,γ,vDF(s,q1,q2)=−γs−c(1−ρ)(q1+q2)+ρ2 E[RDF(D;0,0,s)]+(1−ρ)2 E[RDF(D;q1,q2,s)]+ρ(1−ρ) E[RDF(D;q1,0,s)]+ρ(1−ρ) E[RDF(D;0,q2,s)],P^{DF}_{c,\gamma,v}(s,q_1,q_2) = -\gamma s - c(1-\rho)(q_1+q_2) + \rho^2\,\mathbb E[R^{DF}(D;0,0,s)] + (1-\rho)^2\,\mathbb E[R^{DF}(D;q_1,q_2,s)] + \rho(1-\rho)\,\mathbb E[R^{DF}(D;q_1,0,s)] + \rho(1-\rho)\,\mathbb E[R^{DF}(D;0,q_2,s)],Pc,γ,vDF​(s,q1​,q2​)=−γs−c(1−ρ)(q1​+q2​)+ρ2E[RDF(D;0,0,s)]+(1−ρ)2E[RDF(D;q1​,q2​,s)]+ρ(1−ρ)E[RDF(D;q1​,0,s)]+ρ(1−ρ)E[RDF(D;0,q2​,s)], Pc,γ,vFF(s,q)=−γs−c(1−ρ) q+ρ E[RDF(D;0,0,s)]+(1−ρ) E[RFF(D;q,s)].P^{FF}_{c,\gamma,v}(s,q) = -\gamma s - c(1-\rho)\,q + \rho\,\mathbb E[R^{DF}(D;0,0,s)] + (1-\rho)\,\mathbb E[R^{FF}(D;q,s)].Pc,γ,vFF​(s,q)=−γs−c(1−ρ)q+ρE[RDF(D;0,0,s)]+(1−ρ)E[RFF(D;q,s)].

Optimal values. The two values are suprema taken in the extended reals [−∞,+∞][-\infty,+\infty][−∞,+∞], so either may equal +∞+\infty+∞:

VDF(c,γ,v)=sup⁡s,q1,q2≥0Pc,γ,vDF(s,q1,q2),VFF(c,γ,v)=sup⁡s,q≥0Pc,γ,vFF(s,q).V^{DF}(c,\gamma,v) = \sup_{s,q_1,q_2\ge 0} P^{DF}_{c,\gamma,v}(s,q_1,q_2), \qquad V^{FF}(c,\gamma,v) = \sup_{s,q\ge 0} P^{FF}_{c,\gamma,v}(s,q).VDF(c,γ,v)=s,q1​,q2​≥0sup​Pc,γ,vDF​(s,q1​,q2​),VFF(c,γ,v)=s,q≥0sup​Pc,γ,vFF​(s,q).

Admissible fees and thresholds. A real γ\gammaγ is admissible for (c,v)(c,v)(c,v) when 0≤γ0\le\gamma0≤γ and c<γ+vc<\gamma+vc<γ+v. A real number Γ\GammaΓ is a threshold for (c,v)(c,v)(c,v) when all four of the following hold:

  1. some admissible γ\gammaγ has γ+v≤Γ\gamma+v\le\Gammaγ+v≤Γ;
  2. every admissible γ\gammaγ with γ+v≤Γ\gamma+v\le\Gammaγ+v≤Γ satisfies VFF(c,γ,v)≤VDF(c,γ,v)V^{FF}(c,\gamma,v)\le V^{DF}(c,\gamma,v)VFF(c,γ,v)≤VDF(c,γ,v);
  3. every admissible γ\gammaγ with Γ<γ+v\Gamma<\gamma+vΓ<γ+v satisfies VDF(c,γ,v)≤VFF(c,γ,v)V^{DF}(c,\gamma,v)\le V^{FF}(c,\gamma,v)VDF(c,γ,v)≤VFF(c,γ,v);
  4. some admissible γ\gammaγ with Γ<γ+v\Gamma<\gamma+vΓ<γ+v satisfies the strict inequality VDF(c,γ,v)<VFF(c,γ,v)V^{DF}(c,\gamma,v)< V^{FF}(c,\gamma,v)VDF(c,γ,v)<VFF(c,γ,v).

The threshold is thus compared with the quantity γ+v\gamma+vγ+v, not with γ\gammaγ alone.

The assertion. Assume π2<π1\pi_2<\pi_1π2​<π1​, 0<ρ<10<\rho<10<ρ<1, and a real vvv with 0≤v<π20\le v<\pi_20≤v<π2​; these force 0<π2<π10<\pi_2<\pi_10<π2​<π1​. Then there exists a function Γ:R→R\Gamma:\mathbb R\to\mathbb RΓ:R→R with the following three properties:

  • (a) For every c∈[0,v]c\in[0,v]c∈[0,v], the number Γ(c)\Gamma(c)Γ(c) is a threshold for (c,v)(c,v)(c,v), and it is the only real threshold for (c,v)(c,v)(c,v).
  • (b) For every ccc in the open interval (0,v)(0,v)(0,v), Γ\GammaΓ is differentiable at ccc in the ordinary two-sided sense, with Γ′(c)≤0\Gamma'(c)\le 0Γ′(c)≤0.
  • (c) At each of c=0c=0c=0 and c=vc=vc=v, Γ\GammaΓ has a derivative a≤0a\le 0a≤0 relative to the set [0,v][0,v][0,v], meaning a one-sided derivative using only points of [0,v][0,v][0,v].

The values of Γ\GammaΓ outside [0,v][0,v][0,v] are not constrained by anything in the statement.

Degenerate cases.

  • Bounds on the values. At s=q1=q2=q=0s=q_1=q_2=q=0s=q1​=q2​=q=0 every revenue term is 000, so both profits equal 000. Hence VDFV^{DF}VDF and VFFV^{FF}VFF always lie in [0,+∞][0,+\infty][0,+∞]. Where both equal +∞+\infty+∞, the weak comparisons in clauses 2 and 3 hold, and the strict one in clause 4 fails.
  • The integral convention. For fixed finite decisions the integrands are bounded continuous functions of DDD. So the "non-integrable gives 000" convention does not come into play, even when μ1\mu_1μ1​ or μ2\mu_2μ2​ has infinite mean.
  • Constraints on any threshold. Admissibility forces γ+v>c\gamma+v>cγ+v>c and γ+v≥v\gamma+v\ge vγ+v≥v. So clause 1 makes any threshold satisfy Γ>c\Gamma>cΓ>c and Γ≥v\Gamma\ge vΓ≥v.
  • The case v=0v=0v=0. Then [0,v]={0}[0,v]=\{0\}[0,v]={0} and (0,v)(0,v)(0,v) is empty, so property (b) is vacuous. Property (a) concerns only c=0c=0c=0, where admissibility means γ>0\gamma>0γ>0. Property (c) asks for a derivative relative to the single-point set {0}\{0\}{0}; any number qualifies as such a derivative, so (c) holds with a=0a=0a=0 whatever Γ\GammaΓ is.
  • Both demand laws at 000. The statement places no condition on μ1,μ2\mu_1,\mu_2μ1​,μ2​; in particular it covers both laws concentrated at 000. Then every revenue term vanishes. For c∈[0,v]c\in[0,v]c∈[0,v] and admissible γ\gammaγ, both profits are ≤0\le 0≤0 (because c,γ≥0c,\gamma\ge 0c,γ≥0 and 1−ρ>01-\rho>01−ρ>0) and equal 000 at zero decisions. So VDF=VFF=0V^{DF}=V^{FF}=0VDF=VFF=0 for every admissible γ\gammaγ. Clause 4 would then require 0<00<00<0, so no real number is a threshold for any such (c,v)(c,v)(c,v), yet property (a) asserts that one exists.
Human review
  • Endorsed by Community (Bot) · Oct 29, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Shuze Chen · Oct 29, 2026

    Confirmed by the mission captain (proposal self-audit).

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me